Hiring Intelligence

The Problem

Every hiring team has a spreadsheet. Ours had 307 rows, five per-role tabs, a manually refreshed pivot table, and an interview calendar that broke every time someone rescheduled. Three people were updating three versions of the truth — and we still couldn't answer: "Where are we this week?"

01

No single view across all 6 roles. Every status update meant opening five separate tabs and cross-referencing manually.

02

Interview calendar completely disconnected from the pipeline. Scheduled interviews didn't update candidate stage.

03

No forecasting, no funnel data, no source analysis. No way to know when a role would close or where it was leaking.

Configurable Pipeline

Applied

Shortlisted

Task Sent

Task Recv.

Interview R1

F2F

Offer

Joined

Stages configurable per role — LGS has Mock Call, BD Intern has G-Meet. Each stage carries an SLA; breaches surface on Command Centre automatically.

What Was Built

Command Centre

All 6 roles as live cards — candidate counts per stage, days-open, days-to-close, SLA breach count, and a RAG dot driven by the join-by date. The one screen leadership needs.

Kanban Pipeline

Per-role Kanban with configurable stages. Cards show source, rating, rejection notes. Every edit writes back to Google Sheets. Stage changes trigger pre-filled candidate emails via Gmail API.

Pipeline Insights

Funnel conversion rates, source performance (shortlist % and interview % per source), SLA breach tracker, and a weekly outbound target table with actuals vs goals by source. Built for the Monday TA sync.

AI Forecast

Per-role close date prediction with confidence score. The AI receives funnel conversion rates, source breakdown, stuck candidates, avg scores, and rejection notes — not just headcounts. Suggestions cite your actual data.

Talent Pool

Silver-medal candidates matched to new roles by AI. Soft-reject filter separates "WFH preference" from "portfolio rejection." One click to re-engage; confirmed responses create a new pipeline row with full context.

Interview Log

Per-round logging: interviewer, date, outcome, feedback. History column appends every update. AI and recruiter ratings use the same 1–5 rubric so you can compare First Cut's score with what the interviewer thought.

The Intelligence Layer

"We lose 60% between Shortlisted and Task Sent. That's not a sourcing problem — it's a speed problem. LinkedIn Outbound converts at 22% vs CareerSite at 8%. Shift outbound budget now."

That's what the AI says when it has funnel data. Before, it said "increase outreach volume." The difference is what goes into the prompt.

What the AI receives

Cumulative funnel counts with stage-to-stage conversion rates

Source performance: shortlist % and interview % per source

Candidates stuck beyond SLA — name, stage, days overdue

Avg AI score of shortlisted vs rejected candidates

Last 15 rejection notes verbatim

What the AI produces

Predicted close date with confidence score (0–100%)

3 bottleneck diagnoses citing specific stages and drop-off rates

5 actionable suggestions with named channels and copy angles

Source recommendation: where to focus next week's outbound

Follow-up actions for each stuck candidate by name

What Makes This Different

01

Sheets stays the source of truth

The TA team never changed how they work. Every app edit writes to the same Daily Tracker tab they already maintain. No migration, no retraining. The tool is a better surface on top of a sheet — not a replacement.

02

Dedup across three input sources

Candidates arrive via First Cut (automated screening), manual app entry, and direct sheet edits. A central upsert endpoint matches on LinkedIn URL, email, or name+role before writing. Same candidate, same role — one row, always.

03

One rating scale for AI and recruiters

First Cut scores 0–20. The app maps this to 1–5: 17–20 = ★★★★★, 13–16 = ★★★★, and so on. Recruiters rate on the same rubric. When someone joins, you can compare what the AI called vs what the interviewer said. That's the back-propagation loop — it gets sharper with every hire.

04

Reuse before you re-source

354 candidates screened. Most rejected for soft reasons — WFH preference, notice period, wrong timing. The Talent Pool engine separates these from hard-nos and matches them to new openings. Priya was rejected for LGS because she wanted remote work. The BD Intern role is flexible. She's worth a call before you post again.

Shared Rating Scale

The same 1–5 scale applies to AI scores from First Cut and manual recruiter ratings. Consistent across every candidate, every role, every hire.

★★★★★

Strong Yes
Fast-track

17–20 / 20

★★★★

Yes
Proceed

13–16 / 20

★★★

Maybe
Conditional

8–12 / 20

★★

Weak
Pass w/ reason

4–7 / 20

★

Hard no
Don't reuse

0–3 / 20

Stack

Netlify Cloudflare Worker Google Sheets API Claude Sonnet Gmail API Google OAuth 2.0 Single HTML · no build step @netbramha.com gated